This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: RS6000 fix pr16480
>>>>> Alan Modra writes:
Alan> I should have made it clear in my patch submission: I didn't just
Alan> simply remove the abort to work around a problem. Instead, I looked at
Alan> all the types of address that rs6000_legitimate_address allows. The
Alan> REG_P test is necessary, and I think the rs6000_split_multireg_move code
Alan> now will handle all the addresses that it needs to. (*)
I think we're not communicating. I was not suggesting that the
removal of abort was gratuitous.
My question is whether it truly is okay to follow the codepath
that you changed and leave j=-1. Other paths leave j=-1, but the path you
changed used to assert that it had discovered a REG and would set j or it
had not found a REG and would abort.
As far as I can understand, this would mean that SRC is
(mem (symbol_ref)) or (mem (const (plus (symbol_ref) (const_int))))
and DST is a GPR.
So you patch changes the behavior for LO_SUM, SYMBOL_REF, and
CONST, which previously would have aborted. Now it sets j for the former
and leaves j=-1 for the latter two.
I was not sure from the patch if that is what you intended.
Thanks, David